Group presentation

Frederick, Sofia, Alex

2024-01-24

Data cleaning- test test

  • Cleaning data

  • Tidy data

  • Joining datasets

Feature selection

-   Campbell, M., Hinton, J. D. X. & Anderson, J. R. (2019) A systematic review of the relationship between religion and attitudes toward transgender and gender-variant people. International Journal of Transgenderism, 20:1, 21-38, DOI: 10.1080/15532739.2018.1545149
-   Earle, M. et al. (2021). A multilevel analysis of LGBT (Lesbian, Gay, Bisexual, Transgender) rights support across 77 countries: The role of contact and country laws . British Journal of Social Psychology. doi:10.1111/bjso.12436
-   Flores, A. R. (2015) Attitudes toward transgender rights: perceived knowledge and secondary interpersonal contact, Politics, Groups, and Identities, 3:3, 398-416, DOI: 10.1080/21565503.2015.1050414-
-   Flores, A. R., Brown, T. N. T., & Park, A. S. (2016). Public Support for Transgender Rights: A Twenty-three Country Survey. The Williams Institute at UCLA School of Law. <http://www.jstor.org/stable/resrep34965->
-   Harrison, B.F., Michelson, M.R. (2019) Gender, Masculinity Threat, and Support for Transgender Rights: An Experimental Study. Sex Roles 80, 63–75. <https://doi.org/10.1007/s11199-018-0916-6>
-   Norton, A.T., Herek, G.M. (2013) Heterosexuals’ Attitudes Toward Transgender People: Findings from a National Probability Sample of U.S. Adults. Sex Roles 68, 738–753.

Descriptive Analysis

Binary conception of gender

df_country_1 <- 
  df_descriptive |>
  group_by(country_name) |> 
  summarize(
    prop_gndr_bin = (sum(qc20_multinominal == "Yes") / n())*100, 
    prop_qc19_yes = (sum(qc19 == 1) / n())*100)


cor(df_country_1$prop_gndr_bin, df_country_1$prop_qc19_yes, method = "spearman")
[1] 0.9226601

Religiosity


   Cell Contents
|-------------------------|
|                   Count |
|             Row Percent |
|          Column Percent |
|           Total Percent |
|           Adj Std Resid |
|-------------------------|

Total Observations in Table:  27438 

                       | df_descriptive$qc19_multinominal 
df_descriptive$sd3_cat |        Yes  |         No  | Don't know  |  Row Total | 
-----------------------|------------|------------|------------|------------|
    Atheist | Agnostic |      3860  |      1458  |       485  |      5803  | 
                       |     66.52% |     25.12% |      8.36% |     21.15% | 
                       |     26.69% |     15.04% |     14.79% |            | 
                       |     14.07% |      5.31% |      1.77% |            | 
                       |     23.72  |    -18.32  |     -9.51  |            | 
-----------------------|------------|------------|------------|------------|
              Catholic |      5605  |      4241  |      1352  |     11198  | 
                       |     50.05% |     37.87% |     12.07% |     40.81% | 
                       |     38.75% |     43.74% |     41.22% |            | 
                       |     20.43% |     15.46% |      4.93% |            | 
                       |     -7.32  |      7.30  |      0.51  |            | 
-----------------------|------------|------------|------------|------------|
  Don't know | Refusal |       191  |       175  |       117  |       483  | 
                       |     39.54% |     36.23% |     24.22% |      1.76% | 
                       |      1.32% |      1.81% |      3.57% |            | 
                       |      0.70% |      0.64% |      0.43% |            | 
                       |     -5.85  |      0.42  |      8.39  |            | 
-----------------------|------------|------------|------------|------------|
                Jewish |        34  |        21  |         3  |        58  | 
                       |     58.62% |     36.21% |      5.17% |      0.21% | 
                       |      0.24% |      0.22% |      0.09% |            | 
                       |      0.12% |      0.08% |      0.01% |            | 
                       |      0.90  |      0.14  |     -1.59  |            | 
-----------------------|------------|------------|------------|------------|
                Muslim |       159  |       160  |        73  |       392  | 
                       |     40.56% |     40.82% |     18.62% |      1.43% | 
                       |      1.10% |      1.65% |      2.23% |            | 
                       |      0.58% |      0.58% |      0.27% |            | 
                       |     -4.85  |      2.29  |      4.10  |            | 
-----------------------|------------|------------|------------|------------|
    Orthodox Christian |      1421  |      1959  |       636  |      4016  | 
                       |     35.38% |     48.78% |     15.84% |     14.64% | 
                       |      9.83% |     20.21% |     19.39% |            | 
                       |      5.18% |      7.14% |      2.32% |            | 
                       |    -23.81  |     19.29  |      8.21  |            | 
-----------------------|------------|------------|------------|------------|
                 Other |       730  |       425  |       119  |      1274  | 
                       |     57.30% |     33.36% |      9.34% |      4.64% | 
                       |      5.05% |      4.38% |      3.63% |            | 
                       |      2.66% |      1.55% |      0.43% |            | 
                       |      3.36  |     -1.51  |     -2.94  |            | 
-----------------------|------------|------------|------------|------------|
       Other Christian |       597  |       449  |       137  |      1183  | 
                       |     50.46% |     37.95% |     11.58% |      4.31% | 
                       |      4.13% |      4.63% |      4.18% |            | 
                       |      2.18% |      1.64% |      0.50% |            | 
                       |     -1.58  |      1.93  |     -0.40  |            | 
-----------------------|------------|------------|------------|------------|
            Protestant |      1866  |       807  |       358  |      3031  | 
                       |     61.56% |     26.62% |     11.81% |     11.05% | 
                       |     12.90% |      8.32% |     10.91% |            | 
                       |      6.80% |      2.94% |      1.30% |            | 
                       |     10.35  |    -10.64  |     -0.26  |            | 
-----------------------|------------|------------|------------|------------|
          Column Total |     14463  |      9695  |      3280  |     27438  | 
                       |     52.71% |     35.33% |     11.95% |            | 
-----------------------|------------|------------|------------|------------|

 
Statistics for All Table Factors


Pearson's Chi-squared test 
------------------------------------------------------------
Chi^2 =  1200.107     d.f. =  16     p =  1.412899e-245 


 
       Minimum expected frequency: 6.93345 

Contact with transgender individuals

Age

                     Df  Sum Sq Mean Sq F value Pr(>F)    
qc19_multinominal     2  111437   55719   171.1 <2e-16 ***
Residuals         27435 8936216     326                   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Awareness of discrimination

df_country_2 <-
  df_descriptive |> 
  mutate(qc1_8_ordinal = case_when(
    qc1_8_ordinal == "Very wide spread" ~ "Very or fairly wide spread",
    qc1_8_ordinal == "Fairly wide spread" ~ "Very or fairly wide spread",
    TRUE ~ qc1_8_ordinal
  )) |> 
  group_by(country_name) |> 
  summarize(
    prop_dis_wide = (sum(qc1_8_ordinal == "Very or fairly wide spread") / n())*100, 
    prop_qc19_yes = (sum(qc19 == 1) / n())*100)


cor(df_country_2$prop_dis_wide, df_country_2$prop_qc19_yes, method = "spearman")
[1] 0.4738916

Sexual minority

Support for LGB rights


   Cell Contents
|-------------------------|
|                   Count |
|             Row Percent |
|          Column Percent |
|           Total Percent |
|           Adj Std Resid |
|-------------------------|

Total Observations in Table:  27438 

                              | df_descriptive$qc19_multinominal 
df_descriptive$qc15_1_ordinal |        Yes  |         No  | Don't know  |  Row Total | 
------------------------------|------------|------------|------------|------------|
                Totally agree |      8494  |      1806  |       776  |     11076  | 
                              |     76.69% |     16.31% |      7.01% |     40.37% | 
                              |     58.73% |     18.63% |     23.66% |            | 
                              |     30.96% |      6.58% |      2.83% |            | 
                              |     65.45  |    -54.25  |    -20.79  |            | 
------------------------------|------------|------------|------------|------------|
                Tend to agree |      3949  |      2458  |       948  |      7355  | 
                              |     53.69% |     33.42% |     12.89% |     26.81% | 
                              |     27.30% |     25.35% |     28.90% |            | 
                              |     14.39% |      8.96% |      3.46% |            | 
                              |      1.97  |     -4.02  |      2.89  |            | 
------------------------------|------------|------------|------------|------------|
             Tend to disagree |      1042  |      2297  |       438  |      3777  | 
                              |     27.59% |     60.82% |     11.60% |     13.77% | 
                              |      7.20% |     23.69% |     13.35% |            | 
                              |      3.80% |      8.37% |      1.60% |            | 
                              |    -33.30  |     35.28  |     -0.73  |            | 
------------------------------|------------|------------|------------|------------|
             Totally disagree |       690  |      2636  |       458  |      3784  | 
                              |     18.23% |     69.66% |     12.10% |     13.79% | 
                              |      4.77% |     27.19% |     13.96% |            | 
                              |      2.51% |      9.61% |      1.67% |            | 
                              |    -45.75  |     47.58  |      0.31  |            | 
------------------------------|------------|------------|------------|------------|
                   Don't know |       288  |       498  |       660  |      1446  | 
                              |     19.92% |     34.44% |     45.64% |      5.27% | 
                              |      1.99% |      5.14% |     20.12% |            | 
                              |      1.05% |      1.82% |      2.41% |            | 
                              |    -25.66  |     -0.73  |     40.57  |            | 
------------------------------|------------|------------|------------|------------|
                 Column Total |     14463  |      9695  |      3280  |     27438  | 
                              |     52.71% |     35.33% |     11.95% |            | 
------------------------------|------------|------------|------------|------------|

 
Statistics for All Table Factors


Pearson's Chi-squared test 
------------------------------------------------------------
Chi^2 =  7514.449     d.f. =  8     p =  0 


 
       Minimum expected frequency: 172.8581 
df_country_3 <- 
  df_descriptive |>
  mutate(qc15_1_ordinal = case_when(
    qc15_1_ordinal == "Totally agree" ~ "Totally or tend to agree",
    qc15_1_ordinal == "Tend to agree" ~ "Totally or tend to agree",
    TRUE ~ qc15_1_ordinal
  )) |> 
  group_by(country_name) |> 
  summarize(
    prop_lgb_yes = (sum(qc15_1_ordinal == "Totally or tend to agree") / n())*100, 
    prop_qc19_yes = (sum(qc19 == 1) / n())*100)


cor(df_country_3$prop_lgb_yes, df_country_3$prop_qc19_yes, method = "spearman")
[1] 0.8768473

Statistical Modeling

Dependent Variable (qc19)

It relates to this question:

Do you think that transgender persons should be able to change their civil documents to match their inner gender identity?

We will take the values:

  • 1 for “Yes”

  • 2 for “No”

Getting to the modeling

  • Once we had our variables confirmed from the descriptive analysis we moved to model these variables using a hierarchical (multilevel) logistic regression modeling.
  • We checked for missing data again, and we considered imputing the age_stopped_education variable because of the variables containing missing data, it made the most sense. We tested this out to see if it would help with the significance of the variable in the upcoming models, however failed to make a difference ultimately. The best imputation method seem to be PMM-Imputed distribution.

Check variables with missing data

summary(is.na(Data_final_Ind))
  serialid       country_name    binary_qc19        male        
 Mode :logical   Mode :logical   Mode :logical   Mode :logical  
 FALSE:27438     FALSE:27438     FALSE:24158     FALSE:27438    
                                 TRUE :3280                     
    d11          Cat_age_four    Cat_age_six     Cat_age_seven  
 Mode :logical   Mode :logical   Mode :logical   Mode :logical  
 FALSE:27438     FALSE:27438     FALSE:27438     FALSE:27438    
                                                                
 political_ideology Religion_cat    sd1_7_factor    age_stopped_education
 Mode :logical      Mode :logical   Mode :logical   Mode :logical        
 FALSE:22749        FALSE:27438     FALSE:27438     FALSE:25375          
 TRUE :4689                                         TRUE :2063           
 d60_ordinal       sd2_5         area_type       qc15_1_ordinal 
 Mode :logical   Mode :logical   Mode :logical   Mode :logical  
 FALSE:27059     FALSE:27438     FALSE:27424     FALSE:27438    
 TRUE :379                       TRUE :14                       

 iter imp variable
  1   1  age_stopped_education
  1   2  age_stopped_education
  1   3  age_stopped_education
  1   4  age_stopped_education
  1   5  age_stopped_education
  2   1  age_stopped_education
  2   2  age_stopped_education
  2   3  age_stopped_education
  2   4  age_stopped_education
  2   5  age_stopped_education
  3   1  age_stopped_education
  3   2  age_stopped_education
  3   3  age_stopped_education
  3   4  age_stopped_education
  3   5  age_stopped_education
  4   1  age_stopped_education
  4   2  age_stopped_education
  4   3  age_stopped_education
  4   4  age_stopped_education
  4   5  age_stopped_education
  5   1  age_stopped_education
  5   2  age_stopped_education
  5   3  age_stopped_education
  5   4  age_stopped_education
  5   5  age_stopped_education

 iter imp variable
  1   1  age_stopped_education
  1   2  age_stopped_education
  1   3  age_stopped_education
  1   4  age_stopped_education
  1   5  age_stopped_education
  2   1  age_stopped_education
  2   2  age_stopped_education
  2   3  age_stopped_education
  2   4  age_stopped_education
  2   5  age_stopped_education
  3   1  age_stopped_education
  3   2  age_stopped_education
  3   3  age_stopped_education
  3   4  age_stopped_education
  3   5  age_stopped_education
  4   1  age_stopped_education
  4   2  age_stopped_education
  4   3  age_stopped_education
  4   4  age_stopped_education
  4   5  age_stopped_education
  5   1  age_stopped_education
  5   2  age_stopped_education
  5   3  age_stopped_education
  5   4  age_stopped_education
  5   5  age_stopped_education

 iter imp variable
  1   1  age_stopped_education
  1   2  age_stopped_education
  1   3  age_stopped_education
  1   4  age_stopped_education
  1   5  age_stopped_education
  2   1  age_stopped_education
  2   2  age_stopped_education
  2   3  age_stopped_education
  2   4  age_stopped_education
  2   5  age_stopped_education
  3   1  age_stopped_education
  3   2  age_stopped_education
  3   3  age_stopped_education
  3   4  age_stopped_education
  3   5  age_stopped_education
  4   1  age_stopped_education
  4   2  age_stopped_education
  4   3  age_stopped_education
  4   4  age_stopped_education
  4   5  age_stopped_education
  5   1  age_stopped_education
  5   2  age_stopped_education
  5   3  age_stopped_education
  5   4  age_stopped_education
  5   5  age_stopped_education

 iter imp variable
  1   1  age_stopped_education
  1   2  age_stopped_education
  1   3  age_stopped_education
  1   4  age_stopped_education
  1   5  age_stopped_education
  2   1  age_stopped_education
  2   2  age_stopped_education
  2   3  age_stopped_education
  2   4  age_stopped_education
  2   5  age_stopped_education
  3   1  age_stopped_education
  3   2  age_stopped_education
  3   3  age_stopped_education
  3   4  age_stopped_education
  3   5  age_stopped_education
  4   1  age_stopped_education
  4   2  age_stopped_education
  4   3  age_stopped_education
  4   4  age_stopped_education
  4   5  age_stopped_education
  5   1  age_stopped_education
  5   2  age_stopped_education
  5   3  age_stopped_education
  5   4  age_stopped_education
  5   5  age_stopped_education

Imputation

# Combine the new set of plots into a grid
plot_grid(plotlist = plots, nrow = 3, ncol = 2)

Multi-level analysis

In order to do the multi-leveled analysis we used a generalized linear mixed-effects model (GLMM), where country_name was the variable used as the random effects.

glm_model_2 <- glmer(binary_qc19 ~ male + d11 + I(d11^2) + political_ideology +
               Religion_cat + sd1_7_factor + d60_ordinal + qc15_1_ordinal +
               prop_gndr_bin + prop_dis_wide + Unemployment + (1 | country_name),
               data = Data, family = binomial, 
               control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 100000)))
summary(glm_model_2)
Generalized linear mixed model fit by maximum likelihood (Laplace
  Approximation) [glmerMod]
 Family: binomial  ( logit )
Formula: 
binary_qc19 ~ male + d11 + I(d11^2) + political_ideology + Religion_cat +  
    sd1_7_factor + d60_ordinal + qc15_1_ordinal + prop_gndr_bin +  
    prop_dis_wide + Unemployment + (1 | country_name)
   Data: Data
Control: glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 1e+05))

     AIC      BIC   logLik deviance df.resid 
 20435.5  20704.5 -10183.7  20367.5    20171 

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-6.4039 -0.6105  0.3309  0.5685  5.9159 

Random effects:
 Groups       Name        Variance Std.Dev.
 country_name (Intercept) 0.1296   0.3601  
Number of obs: 20205, groups:  country_name, 28

Fixed effects:
                                    Estimate Std. Error z value Pr(>|z|)    
(Intercept)                        3.234e-01  3.008e-01   1.075  0.28233    
male                              -3.591e-01  3.530e-02 -10.173  < 2e-16 ***
d11                                2.836e-02  5.113e-03   5.547 2.91e-08 ***
I(d11^2)                          -3.173e-04  4.961e-05  -6.396 1.60e-10 ***
political_ideology2                7.860e-02  1.162e-01   0.676  0.49888    
political_ideology3               -2.507e-02  9.270e-02  -0.270  0.78683    
political_ideology4               -1.105e-01  9.138e-02  -1.209  0.22670    
political_ideology5               -1.462e-01  7.872e-02  -1.857  0.06331 .  
political_ideology6               -2.897e-01  8.955e-02  -3.235  0.00122 ** 
political_ideology7               -5.223e-01  8.989e-02  -5.811 6.21e-09 ***
political_ideology8               -5.675e-01  9.178e-02  -6.183 6.28e-10 ***
political_ideology9               -5.556e-01  1.228e-01  -4.524 6.08e-06 ***
political_ideologyRight           -1.360e-01  9.751e-02  -1.395  0.16296    
Religion_catCatholic              -2.401e-01  5.519e-02  -4.350 1.36e-05 ***
Religion_catJewish                 2.438e-01  3.697e-01   0.659  0.50959    
Religion_catMuslim                -7.643e-01  1.584e-01  -4.826 1.39e-06 ***
Religion_catOrthodox Christian    -2.567e-01  9.259e-02  -2.772  0.00557 ** 
Religion_catOther                 -2.329e-01  1.798e-01  -1.295  0.19532    
Religion_catOther Christian       -4.322e-01  9.184e-02  -4.706 2.52e-06 ***
Religion_catOther Religion        -2.094e-01  8.803e-02  -2.379  0.01737 *  
Religion_catProtestant            -1.982e-01  7.172e-02  -2.763  0.00572 ** 
sd1_7_factorNo                    -6.139e-01  6.735e-02  -9.114  < 2e-16 ***
sd1_7_factorRefusal (SPONTANEOUS) -8.757e-01  1.729e-01  -5.064 4.11e-07 ***
sd1_7_factorDon't know            -5.918e-01  1.226e-01  -4.827 1.39e-06 ***
d60_ordinalFrom time to time       1.609e-01  7.497e-02   2.147  0.03180 *  
d60_ordinalAlmost never/ never     3.073e-01  7.238e-02   4.246 2.18e-05 ***
qc15_1_ordinalTend to agree       -8.462e-01  4.518e-02 -18.731  < 2e-16 ***
qc15_1_ordinalTend to disagree    -1.922e+00  5.607e-02 -34.277  < 2e-16 ***
qc15_1_ordinalTotally disagree    -2.359e+00  6.343e-02 -37.196  < 2e-16 ***
qc15_1_ordinalDon't know          -1.745e+00  1.023e-01 -17.060  < 2e-16 ***
prop_gndr_bin                      4.360e+00  5.621e-01   7.758 8.66e-15 ***
prop_dis_wide                     -1.731e+00  7.248e-01  -2.388  0.01693 *  
Unemployment                       6.801e-02  2.574e-02   2.643  0.00822 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
fit warnings:
Some predictor variables are on very different scales: consider rescaling
optimizer (bobyqa) convergence code: 0 (OK)
Model failed to converge with max|grad| = 6.03438 (tol = 0.002, component 1)
Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?
Model is nearly unidentifiable: large eigenvalue ratio
 - Rescale variables?

The model was clearly overfitted. And it suggested we have scaling issue. We fix this and move to remove the non-significant variables.

Generalized linear mixed model fit by maximum likelihood (Laplace
  Approximation) [glmerMod]
 Family: binomial  ( logit )
Formula: 
binary_qc19 ~ male + d11 + I(d11^2) + political_ideology + Religion_cat +  
    sd1_7_factor + d60_ordinal + qc15_1_ordinal + prop_gndr_bin +  
    prop_dis_wide + Unemployment + (1 | country_name)
   Data: Data
Control: glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 1e+05))

     AIC      BIC   logLik deviance df.resid 
 20435.4  20704.5 -10183.7  20367.4    20171 

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-6.4121 -0.6099  0.3306  0.5684  5.9260 

Random effects:
 Groups       Name        Variance Std.Dev.
 country_name (Intercept) 0.1296   0.3601  
Number of obs: 20205, groups:  country_name, 28

Fixed effects:
                                  Estimate Std. Error z value Pr(>|z|)    
(Intercept)                        1.34649    0.29755   4.525 6.03e-06 ***
male                              -0.35962    0.03531 -10.184  < 2e-16 ***
d11                               -0.07900    0.01901  -4.155 3.26e-05 ***
I(d11^2)                          -0.10476    0.01764  -5.940 2.85e-09 ***
political_ideology2                0.07868    0.11625   0.677  0.49854    
political_ideology3               -0.02514    0.09271  -0.271  0.78627    
political_ideology4               -0.11062    0.09140  -1.210  0.22615    
political_ideology5               -0.14639    0.07873  -1.859  0.06297 .  
political_ideology6               -0.29010    0.08956  -3.239  0.00120 ** 
political_ideology7               -0.52306    0.08990  -5.818 5.95e-09 ***
political_ideology8               -0.56835    0.09180  -6.191 5.98e-10 ***
political_ideology9               -0.55644    0.12287  -4.529 5.93e-06 ***
political_ideologyRight           -0.13620    0.09752  -1.397  0.16254    
Religion_catCatholic              -0.24032    0.05520  -4.354 1.34e-05 ***
Religion_catJewish                 0.24420    0.36975   0.660  0.50897    
Religion_catMuslim                -0.76531    0.15838  -4.832 1.35e-06 ***
Religion_catOrthodox Christian    -0.25703    0.09259  -2.776  0.00550 ** 
Religion_catOther                 -0.23311    0.17985  -1.296  0.19493    
Religion_catOther Christian       -0.43283    0.09186  -4.712 2.46e-06 ***
Religion_catOther Religion        -0.20964    0.08805  -2.381  0.01727 *  
Religion_catProtestant            -0.19850    0.07174  -2.767  0.00566 ** 
sd1_7_factorNo                    -0.61463    0.06737  -9.123  < 2e-16 ***
sd1_7_factorRefusal (SPONTANEOUS) -0.87689    0.17296  -5.070 3.98e-07 ***
sd1_7_factorDon't know            -0.59259    0.12263  -4.832 1.35e-06 ***
d60_ordinalFrom time to time       0.16122    0.07498   2.150  0.03153 *  
d60_ordinalAlmost never/ never     0.30780    0.07240   4.252 2.12e-05 ***
qc15_1_ordinalTend to agree       -0.84747    0.04519 -18.754  < 2e-16 ***
qc15_1_ordinalTend to disagree    -1.92485    0.05608 -34.322  < 2e-16 ***
qc15_1_ordinalTotally disagree    -2.36292    0.06345 -37.239  < 2e-16 ***
qc15_1_ordinalDon't know          -1.74740    0.10229 -17.083  < 2e-16 ***
prop_gndr_bin                      4.37191    0.55935   7.816 5.45e-15 ***
prop_dis_wide                     -1.73795    0.72266  -2.405  0.01618 *  
Unemployment                       0.22143    0.08374   2.644  0.00819 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Model (stargazer)

We compared a few models with different transformations and different variables and end up with the best model, to this one.


Results of GLMM Analysis
=============================================================
                                      Dependent variable:    
                                  ---------------------------
                                          binary_qc19        
-------------------------------------------------------------
male                                       -0.360***         
                                            (0.035)          
                                                             
d11                                        -0.079***         
                                            (0.019)          
                                                             
I(d112)                                    -0.105***         
                                            (0.018)          
                                                             
political_ideology2                          0.079           
                                            (0.116)          
                                                             
political_ideology3                         -0.025           
                                            (0.093)          
                                                             
political_ideology4                         -0.111           
                                            (0.091)          
                                                             
political_ideology5                         -0.146*          
                                            (0.079)          
                                                             
political_ideology6                        -0.290***         
                                            (0.090)          
                                                             
political_ideology7                        -0.523***         
                                            (0.090)          
                                                             
political_ideology8                        -0.568***         
                                            (0.092)          
                                                             
political_ideology9                        -0.556***         
                                            (0.123)          
                                                             
political_ideologyRight                     -0.136           
                                            (0.098)          
                                                             
Religion_catCatholic                       -0.240***         
                                            (0.055)          
                                                             
Religion_catJewish                           0.244           
                                            (0.370)          
                                                             
Religion_catMuslim                         -0.765***         
                                            (0.158)          
                                                             
Religion_catOrthodox Christian             -0.257***         
                                            (0.093)          
                                                             
Religion_catOther                           -0.233           
                                            (0.180)          
                                                             
Religion_catOther Christian                -0.433***         
                                            (0.092)          
                                                             
Religion_catOther Religion                 -0.210**          
                                            (0.088)          
                                                             
Religion_catProtestant                     -0.198***         
                                            (0.072)          
                                                             
sd1_7_factorNo                             -0.615***         
                                            (0.067)          
                                                             
sd1_7_factorRefusal (SPONTANEOUS)          -0.877***         
                                            (0.173)          
                                                             
sd1_7_factorDon't know                     -0.593***         
                                            (0.123)          
                                                             
d60_ordinalFrom time to time                0.161**          
                                            (0.075)          
                                                             
d60_ordinalAlmost never/ never             0.308***          
                                            (0.072)          
                                                             
qc15_1_ordinalTend to agree                -0.847***         
                                            (0.045)          
                                                             
qc15_1_ordinalTend to disagree             -1.925***         
                                            (0.056)          
                                                             
qc15_1_ordinalTotally disagree             -2.363***         
                                            (0.063)          
                                                             
qc15_1_ordinalDon't know                   -1.747***         
                                            (0.102)          
                                                             
prop_gndr_bin                              4.372***          
                                            (0.559)          
                                                             
prop_dis_wide                              -1.738**          
                                            (0.723)          
                                                             
Unemployment                               0.221***          
                                            (0.084)          
                                                             
Constant                                   1.346***          
                                            (0.298)          
                                                             
-------------------------------------------------------------
Observations                                20,205           
Log Likelihood                            -10,183.720        
Akaike Inf. Crit.                         20,435.440         
Bayesian Inf. Crit.                       20,704.510         
=============================================================
Note:                             *p<0.1; **p<0.05; ***p<0.01
[1] 20435.44
[1] 20704.51

Key Findings - Demographics and Beliefs

  • Gender Influence: Females show higher odds of support compared to males, potentially reflecting societal gender roles and perceptions.

  • Age Dynamics: Support increases with age up to a point, then declines, suggesting life experiences or generational shifts might influence attitudes toward inclusivity.

  • Political Ideology: Conservative alignment correlates with less support for transgender individuals’ rights to change civil documents, showing a 22.8% decrease in support.

  • Religious Impact: Catholic individuals show 21.2% less support, while Jewish individuals are slightly more supportive, though the latter is very small.

Key Findings - Social and Economic Factors

  • Economic Stability: Individuals with no difficulty paying bills are more supportive, linking economic security with supportive attitudes.

  • Contact with Transgender Individuals: Lack of contact is associated with a 45.5% decrease in support, highlighting the importance of visibility and personal relationships.

  • Sexual Minority and Support: Non-members of sexual minorities are less supportive of transgender rights for document changes.

  • Country-Level Factors: General support for a third gender option and awareness of discrimination correlate with supportive attitudes.

Conclusion

  • Cross-Country Differences: Variations in support for transgender rights across countries are influenced by demographics (gender, age), political and religious affiliations, social determinants, awareness of LGBTI issues, and economic conditions.

Prediction

We aim to build a model to predict support for transgender individuals to change legal documents.

Model = binary_qc19 ~ male + d11 + I(d11^2) + political_ideology + Religion_cat + sd1_7_factor + d60_ordinal + qc15_1_ordinal + Unemployment + prop_gndr_bin + prop_dis_wide

Gradient boosting

Confusion Matrix and Statistics

          Reference
Prediction   No  Yes
       No  1173  601
       Yes  408 1898
                                          
               Accuracy : 0.7527          
                 95% CI : (0.7392, 0.7659)
    No Information Rate : 0.6125          
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.4904          
                                          
 Mcnemar's Test P-Value : 1.499e-09       
                                          
            Sensitivity : 0.7419          
            Specificity : 0.7595          
         Pos Pred Value : 0.6612          
         Neg Pred Value : 0.8231          
             Prevalence : 0.3875          
         Detection Rate : 0.2875          
   Detection Prevalence : 0.4348          
      Balanced Accuracy : 0.7507          
                                          
       'Positive' Class : No              
                                          

Variables Importance

Partial dependence plot

ROC Curve

Final Predictive Model

The selected predictive model for our international analysis is Gradient Boosting (GB).

GB not only has one of the highest Accuracy levels but also achieves the most balanced rates of Specificity and Sensitivity. Such a robust performance profile makes it the optimal choice for our data set.

We found that this model, is the most adept and effective at predicting whether countries will support the right of transgender individuals to change their civil documents.

THANK YOU!